ARCSIM offers new AI consultant services with recruitment of graduate assistant

In recent years, the University of Maine’s Advanced Research Computing, Security, and Information Management (ARCSIM) unit has expanded the services available to UMaine and the University of Maine System. Their most recent expansion introduces new artificial intelligence (AI) and Machine Learning consultation resources. This is possible through the addition of graduate assistant Zafaryab Haider. Haider is pursuing a Ph.D. in Electrical and Computer Engineering with his advisor, Prabuddha Chakraborty, assistant professor of electrical and computer engineering at UMaine, and will be working year-round with ARCSIM as a resource for faculty and students interested in using AI and machine learning in their research. For the next two years, Haider will consult with researchers, connecting them to resources and recommending how they can best leverage AI and machine learning for their work. 

Before coming to UMaine, Haider was an assistant professor at Aligarh Muslim University (AMU) in Aligarh, India where he taught courses on computer networks, data structures, information security, and coding. After seven years of teaching, Haider decided to pursue a Ph.D. in the United States.

Haider’s experience with AI and information security makes him an ideal fit for his position within ARCSIM. While teaching in India, Haider guided projects on AI related to healthcare. He was contacted by the Malaria Lab in AMU to implement an AI model. While this was a science-based application, Haider believes that AI is applicable to any field, as long as one has good datasets and clear prediction goals. Haider explained that any repetitive job can be performed using AI, whether that is inferring information, making predictions, or giving some amount of insight. “It’s a good assistant,” Haider stated. “It can give you a good direction to think about, analyze, and get some information. But you can’t completely trust it or make decisions based on its outcomes.”

As a consultant, Haider is able to work with researchers to help identify the most suitable algorithms and model sizes that can be applied to researchers’ datasets. This variety is mainly due to the advancements in AI, including Large Language Models (LLMs) like Generative Pre-Trained Transformer (GPT) and other options such as Convolutional Neural Networks (CNNs). With LLMs, rather than needing to teach the AI everything, the user fine-tunes the model to their specific data sets, simplifying the process. Similarly, CNNs are highly effective for tasks such as image recognition and can be tailored to specific needs. The range of model sizes available today varies greatly depending on the project’s requirements. Some open models use over 70 billion parameters, which may exceed the computational resources of many universities for training. Current models run by tech giants like Google and others may use 550 billion parameter models, and that number is only growing. There is a tradeoff between size and accuracy. Larger models often promise greater accuracy, while also increase the potential for errors due to their complexity. “Ultimately, the more computational resources you have, the better,” Haider remarked. “And the University is doing a good job of trying to acquire more for its researchers.”

To learn more about how Haider and ARCSIM can support different research projects, click here. To request service for a research project, click here. You can also reach out to ARCSIM at um.arcsim@maine.edu.